Dynflow [DYN(amic work)FLOW] is a workflow engine written in Ruby that allows to:

• keep track of the progress of running process
• run the code asynchronously
• resume the process when something goes wrong, skip some steps when needed
• detect independent parts and run them concurrently
• compose simple actions into more complex scenarios
• extend the workflows from third-party libraries
• keep consistency between local transactional database and external services
• define the input/output interface between the building blocks (planned)
• define rollback for the workflow (planned)
• have multiple workers for distributing the load (planned)

Dynflow doesn't try to choose the best tool for the jobs, as the right tool depends on the context. Instead, it provides interfaces for persistence, transaction layer or executor implementation, giving you the last word in choosing the right one (providing default implementations as well).

• Current status
• How it works
• Glossary
• Related projects

Current status

Dynflow has been under heavy development for several months to be able to support the services orchestration in the Katello and Foreman projects, getting to production-ready state in couple of weeks.

How it works

In traditional workflow engines, you specify a static workflow and then run it with various inputs. Dynflow takes different approach. You specify the inputs and the workflow is generated on the fly. You can either specify the steps explicitly or subscribe one action to another. This is suitable for plugin architecture, where you can't write the whole process on one place.

Dynflow doesn't differentiate between workflow and action. Instead, every action can populate another actions. This allows composing more simpler workflows into a big one.

The whole execution is done in three phases:

1. Plan phase

Construct the execution plan for the workflow. Two mechanisms are used to get the set of actions to be executed:

a. explicit calls of `plan_action` methods in the `plan` method

b. implicit associations: an action A subscribes to an action B,
which means that the action A is executed whenever the action B
occurs.

The output of this phase is a set of actions and their inputs.

2. Run phase

The plan is being executed step by step, calling the run method of an action with corresponding input. The results of every action are written into output attribute.

The run method should be stateless, with all the needed information included in the input from planning phase. This allows us to control the workflow execution: the state of every action can be serialized therefore the workflow itself can be persisted. This makes it easy to recover from failed actions by rerunning it.

3. Finalize phase

Take the results from the execution phase and perform some additional tasks. This is suitable for example for recording the results into database.

Every action can participate in every phase.

Example

One code snippet is worth 1000 words:

# The anatomy of action class

# every action needs to inherit from Dynflow::Action
class Action < Dynflow::Action

  # OPTIONAL: the input format for the execution phase of this action
  # (https://github.com/iNecas/apipie-params for more details.
  # Validations can be performed against this description (turned off
  # for now)
  input_format do
    param :id, Integer
    param :name, String
  end

  # OPTIONAL: every action can produce an output in the execution
  # phase. This allows to describe the output.
  output_format do
    param :uuid, String
  end

  # OPTIONAL: this specifies that this action should be performed when
  # AnotherAction is triggered.
  def self.subscribe
    AnotherAction
  end

  # OPTIONAL: executed during the planning phase. It's possible to
  # specify explicitly the workflow here. By default it schedules just
  # this action.
  def plan(object_1, object_2)
    # +plan_action+ schedules the SubAction to be part of this
    # workflow
    # the +object_1+ is passed to the +SubAction#plan+ method.
    plan_action SubAction, object_1
    # we can specify, where in the workflow this action should be
    # placed, as well as prepare the input.
    plan_self { id: object_2.id, name: object_2.name}
  end

  # OPTIONAL: run the execution part of this action. Transform the
  # data from +input+ to +output+. When not specified, the action is
  # not used in the execution phase.
  def run
    output[:uuid] = "#{input[:name]}-#{input[:id]}"
  end

  # OPTIONAL: finalize the action after the execution phase finishes.
  # in the +input+ and +output+ attributes are available the data from
  # execution phase. in the +outputs+ argument, all the execution
  # phase actions are available, each providing its input and output.
  def finalize
    puts output[:uuid]
  end
end

Every action should be as atomic as possible, providing better granularity when manipulating the process. Since every action can be subscribed by another one, adding new behaviour to an existing workflow is really simple.

The input and output format can be used for defining the interface that other developers can use when extending the workflows.

See the examples directory for more complete examples.

Glossary

• action - building block for the workflows: a Ruby class inherited from Dynflow::Action. Defines code to be run in plan/run/finalize phase. It has defined input and output data.
• execution plan - definition of the workflow: product of the plan phase
• trigger an action - entering the plan phase, starting with the plan method of the action. The execution follows immediately.
• plan_self - converts the arguments of the plan method into action input, that can be accessed from the run/finalize phase.
• plan_action - includes another action into the workflow, passing the arguments into the plan method of the action
• step - execution unit of the action. It represents the action in specific phase (plan step, run step, finalize step).
• flow - definition of the run/finalize phase, holding the information about steps that can run concurrently/in sequence. Part of execution plan.
• executor - service that executes the run and finalize flows based on the execution plan. It can run in the same process as the plan phase or in different process (using the remote executor)
• world - the universe where the Dynflow runs the code: it holds all needed configuration.

Related projects

• Foreman - lifecycle management tool for physical and virtual servers

• Katello - content management plugin for Foreman: integrates couple of REST services for managing the software updates in the infrastructure.

• Foreman-tasks - Foreman plugin providing the tasks management with Dynflow on the back-end

• Dyntask - generic Rails engine providing the tasks management features with Dynflow on the back-end

• Sysflow - set of reusable tools for running system tasks with Dynflow, comes with simple Web-UI for testing it

License

MIT

Authors

Ivan Nečas, Petr Chalupa